Beyond regional greening: Nonlinear GPP trajectories and their environmental associations in eastern coastal China

Coastal ecosystems are highly vulnerable to synergistic environmental pressures, yet traditional linear assessments may hide their short-term dynamic responses. To examine these responses, this study developed an integrated diagnostic framework that combined the Mann–Kendall test, BFAST01, and spatially validated XGBoost–SHAP to characterize GPP trajectories in Shandong Province from 2000 to 2022. The Mann–Kendall analysis showed an extremely significant increasing trend in 69.4% of the pixels. BFAST01 further identified eight trajectory types, among which the monotone increasing trajectory covered 54.5% of the study area. The “Interruption: increasing with negative interruption” type (Type 5) and the “Reversal: from increasing to decreasing” type (Type 7) represented 17.5% and 6.0% of the area, respectively. Approximately 92% of the nonlinear trajectories were significant in at least one segment. Among the Type 5 and Type 7 pixels, the XGBoost classifier achieved a ROC-AUC of 0.87, an average precision of 0.68, and a balanced accuracy of 0.78 in an independent 40-km spatial holdout set. Its classification performance remained stable across repeated spatial partitions and block sizes of 20–60 km. SHAP analysis identified ΔT and ΔP as the leading predictors, with mean absolute SHAP values of 0.74 and 0.73, respectively. Hydroclimatic and topographic variables contributed 60.1% and 24.1% of the summed SHAP importance, while ΔP × ΔT exhibited the strongest pairwise interaction. These findings indicate that regional greening can contain contrasting nonlinear trajectories and that Type 5–Type 7 differentiation is closely associated with coupled hydroclimatic and topographic conditions. The integrated framework provides a spatially robust approach for identifying nonlinear GPP changes and supporting regional vegetation monitoring.